An Empirical Study of Generalization of Backpropagation Using Speed-up Techniques

スポンサーリンク

概要

• 論文の詳細を見る

• Flat spots are known to be one cause for slow training in backpropagation (BP). One proposed technique to eliminate flat spots is the sigmoid prime offset. However, we have found that the sigmoid prime offset can greatly deteriorate generalization of the resulting network. Therefore, we propose techniques not only to eliminate flat spots, but also to retain the generalization ability to be as high as standard BP. Simulations show that the proposed techniques are effective for both speed-up and generalization.

• 一般社団法人情報処理学会の論文
• 1997-10-15

著者

• Perez Jose

  Department Of Computer Science And Systems Engineering Muroran Institute Of Technology

• Suzuki Yukinori

  Department Of Computer Science And Systems Engineering Muroran Institute Of Technology

• SUGIOKA ICHIRO

  Department of Computer Science and Systems Engineering, Muroran Institute of Technology

• Sugioka I

  Muroran Inst. Technol.

• Sugioka Ichiro

  Department Of Computer Science And Systems Engineering Muroran Institute Of Technology

関連論文

• Image Compression Using Vector Quantization with Variable Block Size Division
• An Associative Memory System with Fuzzy Numbers (特集:マルチメディア通信と分散処理)
• Anti-Hepatitis A Virus Antibody Response Elicited in Mice by Different Forms of a Synthetic VP1 Peptide
• An Empirical Study of Generalization of Backpropagation Using Speed-up Techniques

スポンサーリンク